CV - Mai Huong Do
Computer Science PhD candidate (expected graduation: March 2027) specializing in performance and energy analysis of machine learning workloads on multi-node CPU/GPU clusters. Over two years of experience deploying ML workloads on Linux-based hardware, developing an open-source automation framework for distributed execution, collecting CPU/DRAM/GPU power measurements and system telemetry, and analyzing experimental data, with a strong focus on reproducible results. Earlier industry experience in R&D hardware testing and failure diagnosis.
Open to opportunities in Europe and Asia.
Experience
- PhD Researcher, IRIT – University of Toulouse
April 2024 - Present
Toulouse, France · ANR DELIGHT project- Automation framework (FedE-ator, open source): Developed a framework to automate the deployment, execution, and monitoring of machine learning and federated learning experiments across multi-node Linux clusters (Grid’5000 testbed, up to 100 nodes), including CPU/GPU energy monitoring.
- Performance and energy monitoring: Implemented distributed telemetry collection using MPI, integrating CPU/DRAM energy measurements (RAPL), GPU power monitoring (NVML), external power meters, and system metrics from procfs.
- Hardware control: Integrated CPU frequency scaling (Linux cpufreq), GPU clock management, and power-limit controls into the experimental workflow, and implemented custom CPU frequency adjustment mechanisms for DVFS policies.
- Workload benchmarking: Evaluated CNN workloads (LeNet-5, MobileNetV2, ResNet18/50) using Flower and FedScale across different CPU/GPU frequencies, analyzing training performance and energy consumption.
- Performance optimization: Designed and implemented three Dynamic Voltage and Frequency Scaling (DVFS) policies for federated learning, evaluating energy-performance trade-offs and reducing client idle time between training rounds (paper submitted).
- Teaching: Teaching assistant at the University of Toulouse, tutorials and lab sessions on MPI and concurrent programming.
- Master’s Researcher / Research Assistant, ANDA Lab – Soongsil University
August 2021 - January 2024
Seoul, South Korea · IITP / MSIT-funded research- System modeling: Modeled mobile edge computing (MEC) federation systems for Industrial IoT and augmented reality (AR) applications, formulating resource allocation and task offloading problems with energy consumption and latency as optimization objectives.
- Optimization algorithm design: Formulated these problems as Markov Decision Processes (MDPs) and designed actor-critic deep reinforcement learning approaches (DDPG) to optimize energy-delay trade-offs.
- Simulation and implementation: Implemented simulation frameworks in Python to evaluate the proposed algorithms under different system configurations.
- AR workload optimization: Extended the approach to AR applications, modeling video frame chunking and task dependencies across multi-stage processing graphs.
- Radio Unit (RU) System Engineer, VACE Antenna Co., Ltd.
June 2020 - August 2021
Ha Nam, Vietnam · R&D RU Systems- Contributed to the initial development of a Radio Unit project within a newly established R&D team, including setting up hardware test stations and validation workflows.
- Performed hardware testing and failure diagnosis on fully assembled Radio Units, tracing issues to individual subsystems using spectrum analyzers, signal generators, power meters, and oscilloscopes.
- Developed a Python tool to quantify void-area percentages in X-ray images for PCB defect screening and rework decisions.
- Presented test results and design-to-measurement discrepancies in weekly technical reviews with customers and headquarters, supporting decisions on product readiness for mass production.
- Student Intern, Institute of Geophysics – Vietnam Academy of Science and Technology
August 2019 - June 2020
Hanoi, Vietnam- Surveyed antenna design trends, designed microstrip patch antennas for GNSS using HFSS, and produced a working prototype.
Technical Skills
- Performance analysis & telemetry: Workload benchmarking and characterization; performance bottleneck and anomaly investigation; CPU/DRAM energy measurement (Intel RAPL); GPU power monitoring (NVIDIA NVML); external power meters; performance and energy trade-off analysis.
- Linux systems & hardware control: Linux system interfaces (procfs, sysfs); CPU frequency scaling and DVFS; GPU clock and power-limit management; system telemetry and resource monitoring.
- Automation & data analysis: Python, Bash; distributed experiment automation; workload execution and telemetry collection; log parsing; data analysis and visualization (NumPy, Matplotlib); statistical comparison; Git.
- ML/AI workloads: PyTorch, TensorFlow; CNN architectures (ResNet18/50, MobileNetV2, LeNet-5); CUDA GPUs; federated learning frameworks (Flower, FedScale).
- Distributed computing & infrastructure: Linux; multi-node CPU/GPU clusters (Grid’5000); OpenMPI; batch scheduling; Miniconda, Python virtual environments.
- Hardware engineering background: RF hardware testing, debugging, and failure analysis; antenna design (HFSS).
Education
- PhD in Computer Science, University of Toulouse, France
April 2024 - March 2027 (expected)- Thesis: “Exploring the tradeoffs between energy and performance of federated learning algorithms in distributed systems”
- Funded by the ANR DELIGHT project
- Advisors: Prof. Georges Da Costa and Dr. Millian Poquet
- MSc in Information Communication Convergence Technology, Soongsil University, South Korea
August 2021 - January 2024- Thesis: “Energy-delay optimization in mobile edge computing federation systems”
- Funded by IITP & MSIT, Korea
- Advisor: Prof. Yoo Myung Sik
- BEng in Electronics and Telecommunications, Hanoi University of Science and Technology, Vietnam
September 2015 - October 2020- Thesis: “Design of a microstrip antenna for GNSS applications with low noise figure and high gain”
- Advisor: Dr. Lam Hong Thach
Selected Publications & Open Source
- Do, H. M., Poquet, M., & Da Costa, G. (2026). “FedE-ator: A Reproducible Framework for Studying Energy Consumption in Federated Learning Systems.” PECS Workshop, Euro-Par 2026. [Code]
- Do, H. M., Poquet, M., & Da Costa, G. (2026). “Towards Sustainable Federated Learning: A Survey on Energy and Experimental Practices.” Sustainable Computing: Informatics and Systems, 101471. [Artifact]
- Full list of publications (including two IEEE Access articles) on Google Scholar; open-source work on GitHub.
Awards
- Best Undergraduate Graduation Project, Telecommunications Council, HUST (2020)
Languages
- Vietnamese (native), English (working proficiency), French (elementary, self-study)
References
- Prof. Georges Da Costa
- SEPIA Team, IRIT, University of Toulouse
- Email: georges.da-costa@irit.fr
- Prof. Yoo Myung Sik
- Department of Electronic Engineering, Soongsil University, Seoul
- Email: myoo@ssu.ac.kr
